4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LNET
35 #include <linux/ctype.h>
36 #include <linux/log2.h>
37 #include <linux/ktime.h>
38 #include <linux/moduleparam.h>
39 #include <linux/uaccess.h>
41 #include <lnet/lib-lnet.h>
43 #define D_LNI D_CONSOLE
46 * initialize ln_api_mutex statically, since it needs to be used in
47 * discovery_set callback. That module parameter callback can be called
48 * before module init completes. The mutex needs to be ready for use then.
50 struct lnet the_lnet = {
51 .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
52 }; /* THE state of the network */
53 EXPORT_SYMBOL(the_lnet);
55 static char *ip2nets = "";
56 module_param(ip2nets, charp, 0444);
57 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
59 static char *networks = "";
60 module_param(networks, charp, 0444);
61 MODULE_PARM_DESC(networks, "local networks");
63 static char *routes = "";
64 module_param(routes, charp, 0444);
65 MODULE_PARM_DESC(routes, "routes to non-local networks");
67 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
68 module_param(rnet_htable_size, int, 0444);
69 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
71 static int use_tcp_bonding = false;
72 module_param(use_tcp_bonding, int, 0444);
73 MODULE_PARM_DESC(use_tcp_bonding,
74 "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
76 unsigned int lnet_numa_range = 0;
77 module_param(lnet_numa_range, uint, 0444);
78 MODULE_PARM_DESC(lnet_numa_range,
79 "NUMA range to consider during Multi-Rail selection");
82 * lnet_health_sensitivity determines by how much we decrement the health
83 * value on sending error. The value defaults to 100, which means health
84 * interface health is decremented by 100 points every failure.
86 unsigned int lnet_health_sensitivity = 100;
87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
88 #ifdef HAVE_KERNEL_PARAM_OPS
89 static struct kernel_param_ops param_ops_health_sensitivity = {
90 .set = sensitivity_set,
93 #define param_check_health_sensitivity(name, p) \
94 __param_check(name, p, int)
95 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
97 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
98 &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
100 MODULE_PARM_DESC(lnet_health_sensitivity,
101 "Value to decrement the health value by on error");
104 * lnet_recovery_interval determines how often we should perform recovery
105 * on unhealthy interfaces.
107 unsigned int lnet_recovery_interval = 1;
108 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
109 #ifdef HAVE_KERNEL_PARAM_OPS
110 static struct kernel_param_ops param_ops_recovery_interval = {
111 .set = recovery_interval_set,
112 .get = param_get_int,
114 #define param_check_recovery_interval(name, p) \
115 __param_check(name, p, int)
116 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
118 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
119 &lnet_recovery_interval, S_IRUGO|S_IWUSR);
121 MODULE_PARM_DESC(lnet_recovery_interval,
122 "Interval to recover unhealthy interfaces in seconds");
124 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
125 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
127 static struct kernel_param_ops param_ops_interfaces_max = {
129 .get = param_get_int,
132 #define param_check_interfaces_max(name, p) \
133 __param_check(name, p, int)
135 #ifdef HAVE_KERNEL_PARAM_OPS
136 module_param(lnet_interfaces_max, interfaces_max, 0644);
138 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
139 ¶m_ops_interfaces_max, 0644);
141 MODULE_PARM_DESC(lnet_interfaces_max,
142 "Maximum number of interfaces in a node.");
144 unsigned lnet_peer_discovery_disabled = 0;
145 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
147 static struct kernel_param_ops param_ops_discovery_disabled = {
148 .set = discovery_set,
149 .get = param_get_int,
152 #define param_check_discovery_disabled(name, p) \
153 __param_check(name, p, int)
154 #ifdef HAVE_KERNEL_PARAM_OPS
155 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
157 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
158 ¶m_ops_discovery_disabled, 0644);
160 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
161 "Set to 1 to disable peer discovery on this node.");
163 unsigned int lnet_drop_asym_route;
164 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
166 static struct kernel_param_ops param_ops_drop_asym_route = {
167 .set = drop_asym_route_set,
168 .get = param_get_int,
171 #define param_check_drop_asym_route(name, p) \
172 __param_check(name, p, int)
173 #ifdef HAVE_KERNEL_PARAM_OPS
174 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
176 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
177 ¶m_ops_drop_asym_route, 0644);
179 MODULE_PARM_DESC(lnet_drop_asym_route,
180 "Set to 1 to drop asymmetrical route messages.");
182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 10
185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
187 #ifdef HAVE_KERNEL_PARAM_OPS
188 static struct kernel_param_ops param_ops_transaction_timeout = {
189 .set = transaction_to_set,
190 .get = param_get_int,
193 #define param_check_transaction_timeout(name, p) \
194 __param_check(name, p, int)
195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
198 &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
200 MODULE_PARM_DESC(lnet_transaction_timeout,
201 "Maximum number of seconds to wait for a peer response.");
203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 3
204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
206 #ifdef HAVE_KERNEL_PARAM_OPS
207 static struct kernel_param_ops param_ops_retry_count = {
208 .set = retry_count_set,
209 .get = param_get_int,
212 #define param_check_retry_count(name, p) \
213 __param_check(name, p, int)
214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
217 &lnet_retry_count, S_IRUGO|S_IWUSR);
219 MODULE_PARM_DESC(lnet_retry_count,
220 "Maximum number of times to retry transmitting a message");
223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
224 unsigned int lnet_current_net_count;
227 * This sequence number keeps track of how many times DLC was used to
228 * update the local NIs. It is incremented when a NI is added or
229 * removed and checked when sending a message to determine if there is
230 * a need to re-run the selection algorithm. See lnet_select_pathway()
231 * for more details on its usage.
233 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
235 static int lnet_ping(struct lnet_process_id id, signed long timeout,
236 struct lnet_process_id __user *ids, int n_ids);
238 static int lnet_discover(struct lnet_process_id id, __u32 force,
239 struct lnet_process_id __user *ids, int n_ids);
242 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
245 unsigned *sensitivity = (unsigned *)kp->arg;
248 rc = kstrtoul(val, 0, &value);
250 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
255 * The purpose of locking the api_mutex here is to ensure that
256 * the correct value ends up stored properly.
258 mutex_lock(&the_lnet.ln_api_mutex);
260 if (value > LNET_MAX_HEALTH_VALUE) {
261 mutex_unlock(&the_lnet.ln_api_mutex);
262 CERROR("Invalid health value. Maximum: %d value = %lu\n",
263 LNET_MAX_HEALTH_VALUE, value);
268 * if we're turning on health then use the health timeout
271 if (*sensitivity == 0 && value != 0) {
272 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
273 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
275 * if we're turning off health then use the no health timeout
278 } else if (*sensitivity != 0 && value == 0) {
279 lnet_transaction_timeout =
280 LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
281 lnet_retry_count = 0;
284 *sensitivity = value;
286 mutex_unlock(&the_lnet.ln_api_mutex);
292 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
295 unsigned *interval = (unsigned *)kp->arg;
298 rc = kstrtoul(val, 0, &value);
300 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
305 CERROR("lnet_recovery_interval must be at least 1 second\n");
310 * The purpose of locking the api_mutex here is to ensure that
311 * the correct value ends up stored properly.
313 mutex_lock(&the_lnet.ln_api_mutex);
317 mutex_unlock(&the_lnet.ln_api_mutex);
323 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
326 unsigned *discovery = (unsigned *)kp->arg;
328 struct lnet_ping_buffer *pbuf;
330 rc = kstrtoul(val, 0, &value);
332 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
336 value = (value) ? 1 : 0;
339 * The purpose of locking the api_mutex here is to ensure that
340 * the correct value ends up stored properly.
342 mutex_lock(&the_lnet.ln_api_mutex);
344 if (value == *discovery) {
345 mutex_unlock(&the_lnet.ln_api_mutex);
351 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
352 mutex_unlock(&the_lnet.ln_api_mutex);
356 /* tell peers that discovery setting has changed */
357 lnet_net_lock(LNET_LOCK_EX);
358 pbuf = the_lnet.ln_ping_target;
360 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
362 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
363 lnet_net_unlock(LNET_LOCK_EX);
365 lnet_push_update_to_peers(1);
367 mutex_unlock(&the_lnet.ln_api_mutex);
373 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
376 unsigned int *drop_asym_route = (unsigned int *)kp->arg;
379 rc = kstrtoul(val, 0, &value);
381 CERROR("Invalid module parameter value for "
382 "'lnet_drop_asym_route'\n");
387 * The purpose of locking the api_mutex here is to ensure that
388 * the correct value ends up stored properly.
390 mutex_lock(&the_lnet.ln_api_mutex);
392 if (value == *drop_asym_route) {
393 mutex_unlock(&the_lnet.ln_api_mutex);
397 *drop_asym_route = value;
399 mutex_unlock(&the_lnet.ln_api_mutex);
405 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
408 unsigned *transaction_to = (unsigned *)kp->arg;
411 rc = kstrtoul(val, 0, &value);
413 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
418 * The purpose of locking the api_mutex here is to ensure that
419 * the correct value ends up stored properly.
421 mutex_lock(&the_lnet.ln_api_mutex);
423 if (value < lnet_retry_count || value == 0) {
424 mutex_unlock(&the_lnet.ln_api_mutex);
425 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
426 "Has to be greater than lnet_retry_count (%u)\n",
427 value, lnet_retry_count);
431 if (value == *transaction_to) {
432 mutex_unlock(&the_lnet.ln_api_mutex);
436 *transaction_to = value;
437 if (lnet_retry_count == 0)
438 lnet_lnd_timeout = value;
440 lnet_lnd_timeout = value / lnet_retry_count;
442 mutex_unlock(&the_lnet.ln_api_mutex);
448 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
451 unsigned *retry_count = (unsigned *)kp->arg;
454 rc = kstrtoul(val, 0, &value);
456 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
461 * The purpose of locking the api_mutex here is to ensure that
462 * the correct value ends up stored properly.
464 mutex_lock(&the_lnet.ln_api_mutex);
466 if (lnet_health_sensitivity == 0) {
467 mutex_unlock(&the_lnet.ln_api_mutex);
468 CERROR("Can not set retry_count when health feature is turned off\n");
472 if (value > lnet_transaction_timeout) {
473 mutex_unlock(&the_lnet.ln_api_mutex);
474 CERROR("Invalid value for lnet_retry_count (%lu). "
475 "Has to be smaller than lnet_transaction_timeout (%u)\n",
476 value, lnet_transaction_timeout);
480 *retry_count = value;
483 lnet_lnd_timeout = lnet_transaction_timeout;
485 lnet_lnd_timeout = lnet_transaction_timeout / value;
487 mutex_unlock(&the_lnet.ln_api_mutex);
493 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
497 rc = kstrtoint(val, 0, &value);
499 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
503 if (value < LNET_INTERFACES_MIN) {
504 CWARN("max interfaces provided are too small, setting to %d\n",
505 LNET_INTERFACES_MAX_DEFAULT);
506 value = LNET_INTERFACES_MAX_DEFAULT;
509 *(int *)kp->arg = value;
515 lnet_get_routes(void)
521 lnet_get_networks(void)
526 if (*networks != 0 && *ip2nets != 0) {
527 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
528 "'ip2nets' but not both at once\n");
533 rc = lnet_parse_ip2nets(&nets, ip2nets);
534 return (rc == 0) ? nets : NULL;
544 lnet_init_locks(void)
546 spin_lock_init(&the_lnet.ln_eq_wait_lock);
547 spin_lock_init(&the_lnet.ln_msg_resend_lock);
548 init_waitqueue_head(&the_lnet.ln_eq_waitq);
549 init_waitqueue_head(&the_lnet.ln_mt_waitq);
550 mutex_init(&the_lnet.ln_lnd_mutex);
554 lnet_fini_locks(void)
558 struct kmem_cache *lnet_mes_cachep; /* MEs kmem_cache */
559 struct kmem_cache *lnet_small_mds_cachep; /* <= LNET_SMALL_MD_SIZE bytes
563 lnet_descriptor_setup(void)
565 /* create specific kmem_cache for MEs and small MDs (i.e., originally
566 * allocated in <size-xxx> kmem_cache).
568 lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
570 if (!lnet_mes_cachep)
573 lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
574 LNET_SMALL_MD_SIZE, 0, 0,
576 if (!lnet_small_mds_cachep)
583 lnet_descriptor_cleanup(void)
586 if (lnet_small_mds_cachep) {
587 kmem_cache_destroy(lnet_small_mds_cachep);
588 lnet_small_mds_cachep = NULL;
591 if (lnet_mes_cachep) {
592 kmem_cache_destroy(lnet_mes_cachep);
593 lnet_mes_cachep = NULL;
598 lnet_create_remote_nets_table(void)
601 struct list_head *hash;
603 LASSERT(the_lnet.ln_remote_nets_hash == NULL);
604 LASSERT(the_lnet.ln_remote_nets_hbits > 0);
605 LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
607 CERROR("Failed to create remote nets hash table\n");
611 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
612 INIT_LIST_HEAD(&hash[i]);
613 the_lnet.ln_remote_nets_hash = hash;
618 lnet_destroy_remote_nets_table(void)
622 if (the_lnet.ln_remote_nets_hash == NULL)
625 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
626 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
628 LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
629 LNET_REMOTE_NETS_HASH_SIZE *
630 sizeof(the_lnet.ln_remote_nets_hash[0]));
631 the_lnet.ln_remote_nets_hash = NULL;
635 lnet_destroy_locks(void)
637 if (the_lnet.ln_res_lock != NULL) {
638 cfs_percpt_lock_free(the_lnet.ln_res_lock);
639 the_lnet.ln_res_lock = NULL;
642 if (the_lnet.ln_net_lock != NULL) {
643 cfs_percpt_lock_free(the_lnet.ln_net_lock);
644 the_lnet.ln_net_lock = NULL;
651 lnet_create_locks(void)
655 the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
656 if (the_lnet.ln_res_lock == NULL)
659 the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
660 if (the_lnet.ln_net_lock == NULL)
666 lnet_destroy_locks();
670 static void lnet_assert_wire_constants(void)
672 /* Wire protocol assertions generated by 'wirecheck'
673 * running on Linux robert.bartonsoftware.com 2.6.8-1.521
674 * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
675 * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
678 CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
679 CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
680 CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
681 CLASSERT(LNET_MSG_ACK == 0);
682 CLASSERT(LNET_MSG_PUT == 1);
683 CLASSERT(LNET_MSG_GET == 2);
684 CLASSERT(LNET_MSG_REPLY == 3);
685 CLASSERT(LNET_MSG_HELLO == 4);
687 /* Checks for struct lnet_handle_wire */
688 CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
689 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
690 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
691 CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
692 CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
694 /* Checks for struct struct lnet_magicversion */
695 CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
696 CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
697 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
698 CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
699 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
700 CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
701 CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
703 /* Checks for struct struct lnet_hdr */
704 CLASSERT((int)sizeof(struct lnet_hdr) == 72);
705 CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
706 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
707 CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
708 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
709 CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
710 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
711 CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
712 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
713 CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
714 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
715 CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
716 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
717 CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
718 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
721 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
722 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
723 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
724 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
725 CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
726 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
729 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
730 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
731 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
732 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
733 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
734 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
735 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
736 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
737 CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
738 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
741 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
742 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
743 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
744 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
745 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
746 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
747 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
748 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
749 CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
750 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
753 CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
754 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
757 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
758 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
759 CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
760 CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
762 /* Checks for struct lnet_ni_status and related constants */
763 CLASSERT(LNET_NI_STATUS_INVALID == 0x00000000);
764 CLASSERT(LNET_NI_STATUS_UP == 0x15aac0de);
765 CLASSERT(LNET_NI_STATUS_DOWN == 0xdeadface);
767 /* Checks for struct lnet_ni_status */
768 CLASSERT((int)sizeof(struct lnet_ni_status) == 16);
769 CLASSERT((int)offsetof(struct lnet_ni_status, ns_nid) == 0);
770 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) == 8);
771 CLASSERT((int)offsetof(struct lnet_ni_status, ns_status) == 8);
772 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_status) == 4);
773 CLASSERT((int)offsetof(struct lnet_ni_status, ns_unused) == 12);
774 CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) == 4);
776 /* Checks for struct lnet_ping_info and related constants */
777 CLASSERT(LNET_PROTO_PING_MAGIC == 0x70696E67);
778 CLASSERT(LNET_PING_FEAT_INVAL == 0);
779 CLASSERT(LNET_PING_FEAT_BASE == 1);
780 CLASSERT(LNET_PING_FEAT_NI_STATUS == 2);
781 CLASSERT(LNET_PING_FEAT_RTE_DISABLED == 4);
782 CLASSERT(LNET_PING_FEAT_MULTI_RAIL == 8);
783 CLASSERT(LNET_PING_FEAT_DISCOVERY == 16);
784 CLASSERT(LNET_PING_FEAT_BITS == 31);
786 /* Checks for struct lnet_ping_info */
787 CLASSERT((int)sizeof(struct lnet_ping_info) == 16);
788 CLASSERT((int)offsetof(struct lnet_ping_info, pi_magic) == 0);
789 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) == 4);
790 CLASSERT((int)offsetof(struct lnet_ping_info, pi_features) == 4);
791 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_features) == 4);
792 CLASSERT((int)offsetof(struct lnet_ping_info, pi_pid) == 8);
793 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) == 4);
794 CLASSERT((int)offsetof(struct lnet_ping_info, pi_nnis) == 12);
795 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) == 4);
796 CLASSERT((int)offsetof(struct lnet_ping_info, pi_ni) == 16);
797 CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) == 0);
800 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
802 struct lnet_lnd *lnd;
803 struct list_head *tmp;
805 /* holding lnd mutex */
806 list_for_each(tmp, &the_lnet.ln_lnds) {
807 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
809 if (lnd->lnd_type == type)
816 lnet_get_lnd_timeout(void)
818 return lnet_lnd_timeout;
820 EXPORT_SYMBOL(lnet_get_lnd_timeout);
823 lnet_register_lnd(struct lnet_lnd *lnd)
825 mutex_lock(&the_lnet.ln_lnd_mutex);
827 LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
828 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
830 list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
831 lnd->lnd_refcount = 0;
833 CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
835 mutex_unlock(&the_lnet.ln_lnd_mutex);
837 EXPORT_SYMBOL(lnet_register_lnd);
840 lnet_unregister_lnd(struct lnet_lnd *lnd)
842 mutex_lock(&the_lnet.ln_lnd_mutex);
844 LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
845 LASSERT(lnd->lnd_refcount == 0);
847 list_del(&lnd->lnd_list);
848 CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
850 mutex_unlock(&the_lnet.ln_lnd_mutex);
852 EXPORT_SYMBOL(lnet_unregister_lnd);
855 lnet_counters_get_common(struct lnet_counters_common *common)
857 struct lnet_counters *ctr;
860 memset(common, 0, sizeof(*common));
862 lnet_net_lock(LNET_LOCK_EX);
864 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
865 common->lcc_msgs_max += ctr->lct_common.lcc_msgs_max;
866 common->lcc_msgs_alloc += ctr->lct_common.lcc_msgs_alloc;
867 common->lcc_errors += ctr->lct_common.lcc_errors;
868 common->lcc_send_count += ctr->lct_common.lcc_send_count;
869 common->lcc_recv_count += ctr->lct_common.lcc_recv_count;
870 common->lcc_route_count += ctr->lct_common.lcc_route_count;
871 common->lcc_drop_count += ctr->lct_common.lcc_drop_count;
872 common->lcc_send_length += ctr->lct_common.lcc_send_length;
873 common->lcc_recv_length += ctr->lct_common.lcc_recv_length;
874 common->lcc_route_length += ctr->lct_common.lcc_route_length;
875 common->lcc_drop_length += ctr->lct_common.lcc_drop_length;
877 lnet_net_unlock(LNET_LOCK_EX);
879 EXPORT_SYMBOL(lnet_counters_get_common);
882 lnet_counters_get(struct lnet_counters *counters)
884 struct lnet_counters *ctr;
885 struct lnet_counters_health *health = &counters->lct_health;
888 memset(counters, 0, sizeof(*counters));
890 lnet_counters_get_common(&counters->lct_common);
892 lnet_net_lock(LNET_LOCK_EX);
894 cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
895 health->lch_rst_alloc += ctr->lct_health.lch_rst_alloc;
896 health->lch_resend_count += ctr->lct_health.lch_resend_count;
897 health->lch_response_timeout_count +=
898 ctr->lct_health.lch_response_timeout_count;
899 health->lch_local_interrupt_count +=
900 ctr->lct_health.lch_local_interrupt_count;
901 health->lch_local_dropped_count +=
902 ctr->lct_health.lch_local_dropped_count;
903 health->lch_local_aborted_count +=
904 ctr->lct_health.lch_local_aborted_count;
905 health->lch_local_no_route_count +=
906 ctr->lct_health.lch_local_no_route_count;
907 health->lch_local_timeout_count +=
908 ctr->lct_health.lch_local_timeout_count;
909 health->lch_local_error_count +=
910 ctr->lct_health.lch_local_error_count;
911 health->lch_remote_dropped_count +=
912 ctr->lct_health.lch_remote_dropped_count;
913 health->lch_remote_error_count +=
914 ctr->lct_health.lch_remote_error_count;
915 health->lch_remote_timeout_count +=
916 ctr->lct_health.lch_remote_timeout_count;
917 health->lch_network_timeout_count +=
918 ctr->lct_health.lch_network_timeout_count;
920 lnet_net_unlock(LNET_LOCK_EX);
922 EXPORT_SYMBOL(lnet_counters_get);
925 lnet_counters_reset(void)
927 struct lnet_counters *counters;
930 lnet_net_lock(LNET_LOCK_EX);
932 cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
933 memset(counters, 0, sizeof(struct lnet_counters));
935 lnet_net_unlock(LNET_LOCK_EX);
939 lnet_res_type2str(int type)
944 case LNET_COOKIE_TYPE_MD:
946 case LNET_COOKIE_TYPE_ME:
948 case LNET_COOKIE_TYPE_EQ:
954 lnet_res_container_cleanup(struct lnet_res_container *rec)
958 if (rec->rec_type == 0) /* not set yet, it's uninitialized */
961 while (!list_empty(&rec->rec_active)) {
962 struct list_head *e = rec->rec_active.next;
965 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
966 lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
968 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
969 lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
971 } else { /* NB: Active MEs should be attached on portals */
978 /* Found alive MD/ME/EQ, user really should unlink/free
979 * all of them before finalize LNet, but if someone didn't,
980 * we have to recycle garbage for him */
981 CERROR("%d active elements on exit of %s container\n",
982 count, lnet_res_type2str(rec->rec_type));
985 if (rec->rec_lh_hash != NULL) {
986 LIBCFS_FREE(rec->rec_lh_hash,
987 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
988 rec->rec_lh_hash = NULL;
991 rec->rec_type = 0; /* mark it as finalized */
995 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1000 LASSERT(rec->rec_type == 0);
1002 rec->rec_type = type;
1003 INIT_LIST_HEAD(&rec->rec_active);
1005 rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1007 /* Arbitrary choice of hash table size */
1008 LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1009 LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1010 if (rec->rec_lh_hash == NULL) {
1015 for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1016 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1021 CERROR("Failed to setup %s resource container\n",
1022 lnet_res_type2str(type));
1023 lnet_res_container_cleanup(rec);
1028 lnet_res_containers_destroy(struct lnet_res_container **recs)
1030 struct lnet_res_container *rec;
1033 cfs_percpt_for_each(rec, i, recs)
1034 lnet_res_container_cleanup(rec);
1036 cfs_percpt_free(recs);
1039 static struct lnet_res_container **
1040 lnet_res_containers_create(int type)
1042 struct lnet_res_container **recs;
1043 struct lnet_res_container *rec;
1047 recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1049 CERROR("Failed to allocate %s resource containers\n",
1050 lnet_res_type2str(type));
1054 cfs_percpt_for_each(rec, i, recs) {
1055 rc = lnet_res_container_setup(rec, i, type);
1057 lnet_res_containers_destroy(recs);
1065 struct lnet_libhandle *
1066 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1068 /* ALWAYS called with lnet_res_lock held */
1069 struct list_head *head;
1070 struct lnet_libhandle *lh;
1073 if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1076 hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1077 head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1079 list_for_each_entry(lh, head, lh_hash_chain) {
1080 if (lh->lh_cookie == cookie)
1088 lnet_res_lh_initialize(struct lnet_res_container *rec,
1089 struct lnet_libhandle *lh)
1091 /* ALWAYS called with lnet_res_lock held */
1092 unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1095 lh->lh_cookie = rec->rec_lh_cookie;
1096 rec->rec_lh_cookie += 1 << ibits;
1098 hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1100 list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1104 lnet_create_array_of_queues(void)
1106 struct list_head **qs;
1107 struct list_head *q;
1110 qs = cfs_percpt_alloc(lnet_cpt_table(),
1111 sizeof(struct list_head));
1113 CERROR("Failed to allocate queues\n");
1117 cfs_percpt_for_each(q, i, qs)
1123 static int lnet_unprepare(void);
1126 lnet_prepare(lnet_pid_t requested_pid)
1128 /* Prepare to bring up the network */
1129 struct lnet_res_container **recs;
1132 if (requested_pid == LNET_PID_ANY) {
1133 /* Don't instantiate LNET just for me */
1137 LASSERT(the_lnet.ln_refcount == 0);
1139 the_lnet.ln_routing = 0;
1141 LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1142 the_lnet.ln_pid = requested_pid;
1144 INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1145 INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1146 INIT_LIST_HEAD(&the_lnet.ln_nets);
1147 INIT_LIST_HEAD(&the_lnet.ln_routers);
1148 INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1149 INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1150 INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1151 INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1152 INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1153 INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1154 INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1155 init_waitqueue_head(&the_lnet.ln_dc_waitq);
1156 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1157 init_completion(&the_lnet.ln_started);
1159 rc = lnet_descriptor_setup();
1163 rc = lnet_create_remote_nets_table();
1168 * NB the interface cookie in wire handles guards against delayed
1169 * replies and ACKs appearing valid after reboot.
1171 the_lnet.ln_interface_cookie = ktime_get_real_ns();
1173 the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1174 sizeof(struct lnet_counters));
1175 if (the_lnet.ln_counters == NULL) {
1176 CERROR("Failed to allocate counters for LNet\n");
1181 rc = lnet_peer_tables_create();
1185 rc = lnet_msg_containers_create();
1189 rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1190 LNET_COOKIE_TYPE_EQ);
1194 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1200 the_lnet.ln_me_containers = recs;
1202 recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1208 the_lnet.ln_md_containers = recs;
1210 rc = lnet_portals_create();
1212 CERROR("Failed to create portals for LNet: %d\n", rc);
1216 the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1217 if (!the_lnet.ln_mt_zombie_rstqs) {
1230 lnet_unprepare (void)
1234 /* NB no LNET_LOCK since this is the last reference. All LND instances
1235 * have shut down already, so it is safe to unlink and free all
1236 * descriptors, even those that appear committed to a network op (eg MD
1237 * with non-zero pending count) */
1239 lnet_fail_nid(LNET_NID_ANY, 0);
1241 LASSERT(the_lnet.ln_refcount == 0);
1242 LASSERT(list_empty(&the_lnet.ln_test_peers));
1243 LASSERT(list_empty(&the_lnet.ln_nets));
1245 if (the_lnet.ln_mt_zombie_rstqs) {
1246 lnet_clean_zombie_rstqs();
1247 the_lnet.ln_mt_zombie_rstqs = NULL;
1250 if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1251 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1252 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1256 lnet_portals_destroy();
1258 if (the_lnet.ln_md_containers != NULL) {
1259 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1260 the_lnet.ln_md_containers = NULL;
1263 if (the_lnet.ln_me_containers != NULL) {
1264 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1265 the_lnet.ln_me_containers = NULL;
1268 lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1270 lnet_msg_containers_destroy();
1272 lnet_rtrpools_free(0);
1274 if (the_lnet.ln_counters != NULL) {
1275 cfs_percpt_free(the_lnet.ln_counters);
1276 the_lnet.ln_counters = NULL;
1278 lnet_destroy_remote_nets_table();
1279 lnet_descriptor_cleanup();
1285 lnet_net2ni_locked(__u32 net_id, int cpt)
1288 struct lnet_net *net;
1290 LASSERT(cpt != LNET_LOCK_EX);
1292 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1293 if (net->net_id == net_id) {
1294 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1304 lnet_net2ni_addref(__u32 net)
1309 ni = lnet_net2ni_locked(net, 0);
1311 lnet_ni_addref_locked(ni, 0);
1316 EXPORT_SYMBOL(lnet_net2ni_addref);
1319 lnet_get_net_locked(__u32 net_id)
1321 struct lnet_net *net;
1323 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1324 if (net->net_id == net_id)
1332 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1337 LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1342 val = hash_long(key, LNET_CPT_BITS);
1343 /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1347 return (unsigned int)(key + val + (val >> 1)) % number;
1351 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1353 struct lnet_net *net;
1355 /* must called with hold of lnet_net_lock */
1356 if (LNET_CPT_NUMBER == 1)
1357 return 0; /* the only one */
1360 * If NI is provided then use the CPT identified in the NI cpt
1361 * list if one exists. If one doesn't exist, then that NI is
1362 * associated with all CPTs and it follows that the net it belongs
1363 * to is implicitly associated with all CPTs, so just hash the nid
1367 if (ni->ni_cpts != NULL)
1368 return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1371 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1374 /* no NI provided so look at the net */
1375 net = lnet_get_net_locked(LNET_NIDNET(nid));
1377 if (net != NULL && net->net_cpts != NULL) {
1378 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1381 return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1385 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1390 if (LNET_CPT_NUMBER == 1)
1391 return 0; /* the only one */
1393 cpt = lnet_net_lock_current();
1395 cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1397 lnet_net_unlock(cpt);
1401 EXPORT_SYMBOL(lnet_cpt_of_nid);
1404 lnet_islocalnet_locked(__u32 net_id)
1406 struct lnet_net *net;
1409 net = lnet_get_net_locked(net_id);
1411 local = net != NULL;
1417 lnet_islocalnet(__u32 net_id)
1422 cpt = lnet_net_lock_current();
1424 local = lnet_islocalnet_locked(net_id);
1426 lnet_net_unlock(cpt);
1432 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1434 struct lnet_net *net;
1437 LASSERT(cpt != LNET_LOCK_EX);
1439 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1440 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1441 if (ni->ni_nid == nid)
1450 lnet_nid2ni_addref(lnet_nid_t nid)
1455 ni = lnet_nid2ni_locked(nid, 0);
1457 lnet_ni_addref_locked(ni, 0);
1462 EXPORT_SYMBOL(lnet_nid2ni_addref);
1465 lnet_islocalnid(lnet_nid_t nid)
1470 cpt = lnet_net_lock_current();
1471 ni = lnet_nid2ni_locked(nid, cpt);
1472 lnet_net_unlock(cpt);
1478 lnet_count_acceptor_nets(void)
1480 /* Return the # of NIs that need the acceptor. */
1482 struct lnet_net *net;
1485 cpt = lnet_net_lock_current();
1486 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1487 /* all socklnd type networks should have the acceptor
1489 if (net->net_lnd->lnd_accept != NULL)
1493 lnet_net_unlock(cpt);
1498 struct lnet_ping_buffer *
1499 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1501 struct lnet_ping_buffer *pbuf;
1503 LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1505 pbuf->pb_nnis = nnis;
1506 atomic_set(&pbuf->pb_refcnt, 1);
1513 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1515 LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1516 LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1519 static struct lnet_ping_buffer *
1520 lnet_ping_target_create(int nnis)
1522 struct lnet_ping_buffer *pbuf;
1524 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1526 CERROR("Can't allocate ping source [%d]\n", nnis);
1530 pbuf->pb_info.pi_nnis = nnis;
1531 pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1532 pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1533 pbuf->pb_info.pi_features =
1534 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1540 lnet_get_net_ni_count_locked(struct lnet_net *net)
1545 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1552 lnet_get_net_ni_count_pre(struct lnet_net *net)
1557 list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1564 lnet_get_ni_count(void)
1567 struct lnet_net *net;
1572 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1573 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1583 lnet_get_net_count(void)
1585 struct lnet_net *net;
1590 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1600 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1602 struct lnet_ni_status *stat;
1606 __swab32s(&pbuf->pb_info.pi_magic);
1607 __swab32s(&pbuf->pb_info.pi_features);
1608 __swab32s(&pbuf->pb_info.pi_pid);
1609 __swab32s(&pbuf->pb_info.pi_nnis);
1610 nnis = pbuf->pb_info.pi_nnis;
1611 if (nnis > pbuf->pb_nnis)
1612 nnis = pbuf->pb_nnis;
1613 for (i = 0; i < nnis; i++) {
1614 stat = &pbuf->pb_info.pi_ni[i];
1615 __swab64s(&stat->ns_nid);
1616 __swab32s(&stat->ns_status);
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1626 if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1628 if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1630 /* Loopback is guaranteed to be present */
1631 if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1633 if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1639 lnet_ping_target_destroy(void)
1641 struct lnet_net *net;
1644 lnet_net_lock(LNET_LOCK_EX);
1646 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1649 ni->ni_status = NULL;
1654 lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655 the_lnet.ln_ping_target = NULL;
1657 lnet_net_unlock(LNET_LOCK_EX);
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1663 struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1665 if (event->unlinked)
1666 lnet_ping_buffer_decref(pbuf);
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671 struct lnet_handle_md *ping_mdh,
1672 int ni_count, bool set_eq)
1674 struct lnet_process_id id = {
1675 .nid = LNET_NID_ANY,
1678 struct lnet_handle_me me_handle;
1679 struct lnet_md md = { NULL };
1683 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1684 &the_lnet.ln_ping_target_eq);
1686 CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1691 *ppbuf = lnet_ping_target_create(ni_count);
1692 if (*ppbuf == NULL) {
1697 /* Ping target ME/MD */
1698 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1699 LNET_PROTO_PING_MATCHBITS, 0,
1700 LNET_UNLINK, LNET_INS_AFTER,
1703 CERROR("Can't create ping target ME: %d\n", rc);
1704 goto fail_decref_ping_buffer;
1707 /* initialize md content */
1708 md.start = &(*ppbuf)->pb_info;
1709 md.length = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1710 md.threshold = LNET_MD_THRESH_INF;
1712 md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1713 LNET_MD_MANAGE_REMOTE;
1714 md.eq_handle = the_lnet.ln_ping_target_eq;
1715 md.user_ptr = *ppbuf;
1717 rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1719 CERROR("Can't attach ping target MD: %d\n", rc);
1720 goto fail_unlink_ping_me;
1722 lnet_ping_buffer_addref(*ppbuf);
1726 fail_unlink_ping_me:
1727 rc2 = LNetMEUnlink(me_handle);
1729 fail_decref_ping_buffer:
1730 LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1731 lnet_ping_buffer_decref(*ppbuf);
1735 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1742 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1743 struct lnet_handle_md *ping_mdh)
1745 sigset_t blocked = cfs_block_allsigs();
1747 LNetMDUnlink(*ping_mdh);
1748 LNetInvalidateMDHandle(ping_mdh);
1750 /* NB the MD could be busy; this just starts the unlink */
1751 while (lnet_ping_buffer_numref(pbuf) > 1) {
1752 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1753 set_current_state(TASK_UNINTERRUPTIBLE);
1754 schedule_timeout(cfs_time_seconds(1));
1757 cfs_restore_sigs(blocked);
1761 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1764 struct lnet_net *net;
1765 struct lnet_ni_status *ns;
1770 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1771 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1772 LASSERT(i < pbuf->pb_nnis);
1774 ns = &pbuf->pb_info.pi_ni[i];
1776 ns->ns_nid = ni->ni_nid;
1779 ns->ns_status = (ni->ni_status != NULL) ?
1780 ni->ni_status->ns_status :
1789 * We (ab)use the ns_status of the loopback interface to
1790 * transmit the sequence number. The first interface listed
1791 * must be the loopback interface.
1793 rc = lnet_ping_info_validate(&pbuf->pb_info);
1795 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1798 LNET_PING_BUFFER_SEQNO(pbuf) =
1799 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1803 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1804 struct lnet_handle_md ping_mdh)
1806 struct lnet_ping_buffer *old_pbuf = NULL;
1807 struct lnet_handle_md old_ping_md;
1809 /* switch the NIs to point to the new ping info created */
1810 lnet_net_lock(LNET_LOCK_EX);
1812 if (!the_lnet.ln_routing)
1813 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1814 if (!lnet_peer_discovery_disabled)
1815 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1817 /* Ensure only known feature bits have been set. */
1818 LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1819 LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1821 lnet_ping_target_install_locked(pbuf);
1823 if (the_lnet.ln_ping_target) {
1824 old_pbuf = the_lnet.ln_ping_target;
1825 old_ping_md = the_lnet.ln_ping_target_md;
1827 the_lnet.ln_ping_target_md = ping_mdh;
1828 the_lnet.ln_ping_target = pbuf;
1830 lnet_net_unlock(LNET_LOCK_EX);
1833 /* unlink and free the old ping info */
1834 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1835 lnet_ping_buffer_decref(old_pbuf);
1838 lnet_push_update_to_peers(0);
1842 lnet_ping_target_fini(void)
1846 lnet_ping_md_unlink(the_lnet.ln_ping_target,
1847 &the_lnet.ln_ping_target_md);
1849 rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1852 lnet_ping_target_destroy();
1855 /* Resize the push target. */
1856 int lnet_push_target_resize(void)
1858 struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1859 struct lnet_md md = { NULL };
1860 struct lnet_handle_me meh;
1861 struct lnet_handle_md mdh;
1862 struct lnet_handle_md old_mdh;
1863 struct lnet_ping_buffer *pbuf;
1864 struct lnet_ping_buffer *old_pbuf;
1865 int nnis = the_lnet.ln_push_target_nnis;
1873 pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1879 rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1880 LNET_PROTO_PING_MATCHBITS, 0,
1881 LNET_UNLINK, LNET_INS_AFTER,
1884 CERROR("Can't create push target ME: %d\n", rc);
1885 goto fail_decref_pbuf;
1888 /* initialize md content */
1889 md.start = &pbuf->pb_info;
1890 md.length = LNET_PING_INFO_SIZE(nnis);
1891 md.threshold = LNET_MD_THRESH_INF;
1893 md.options = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1894 LNET_MD_MANAGE_REMOTE;
1896 md.eq_handle = the_lnet.ln_push_target_eq;
1898 rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1900 CERROR("Can't attach push MD: %d\n", rc);
1901 goto fail_unlink_meh;
1903 lnet_ping_buffer_addref(pbuf);
1905 lnet_net_lock(LNET_LOCK_EX);
1906 old_pbuf = the_lnet.ln_push_target;
1907 old_mdh = the_lnet.ln_push_target_md;
1908 the_lnet.ln_push_target = pbuf;
1909 the_lnet.ln_push_target_md = mdh;
1910 lnet_net_unlock(LNET_LOCK_EX);
1913 LNetMDUnlink(old_mdh);
1914 lnet_ping_buffer_decref(old_pbuf);
1917 if (nnis < the_lnet.ln_push_target_nnis)
1920 CDEBUG(D_NET, "nnis %d success\n", nnis);
1927 lnet_ping_buffer_decref(pbuf);
1929 CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1933 static void lnet_push_target_event_handler(struct lnet_event *ev)
1935 struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1937 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1938 lnet_swap_pinginfo(pbuf);
1940 lnet_peer_push_event(ev);
1942 lnet_ping_buffer_decref(pbuf);
1945 /* Initialize the push target. */
1946 static int lnet_push_target_init(void)
1950 if (the_lnet.ln_push_target)
1953 rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1954 &the_lnet.ln_push_target_eq);
1956 CERROR("Can't allocated push target EQ: %d\n", rc);
1960 /* Start at the required minimum, we'll enlarge if required. */
1961 the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1963 rc = lnet_push_target_resize();
1966 LNetEQFree(the_lnet.ln_push_target_eq);
1967 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1973 /* Clean up the push target. */
1974 static void lnet_push_target_fini(void)
1976 if (!the_lnet.ln_push_target)
1979 /* Unlink and invalidate to prevent new references. */
1980 LNetMDUnlink(the_lnet.ln_push_target_md);
1981 LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1983 /* Wait for the unlink to complete. */
1984 while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1985 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1986 set_current_state(TASK_UNINTERRUPTIBLE);
1987 schedule_timeout(cfs_time_seconds(1));
1990 lnet_ping_buffer_decref(the_lnet.ln_push_target);
1991 the_lnet.ln_push_target = NULL;
1992 the_lnet.ln_push_target_nnis = 0;
1994 LNetEQFree(the_lnet.ln_push_target_eq);
1995 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1999 lnet_ni_tq_credits(struct lnet_ni *ni)
2003 LASSERT(ni->ni_ncpts >= 1);
2005 if (ni->ni_ncpts == 1)
2006 return ni->ni_net->net_tunables.lct_max_tx_credits;
2008 credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2009 credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2010 credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2016 lnet_ni_unlink_locked(struct lnet_ni *ni)
2018 /* move it to zombie list and nobody can find it anymore */
2019 LASSERT(!list_empty(&ni->ni_netlist));
2020 list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2021 lnet_ni_decref_locked(ni, 0);
2025 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2030 struct list_head *zombie_list = &net->net_ni_zombie;
2033 * Now wait for the NIs I just nuked to show up on the zombie
2034 * list and shut them down in guaranteed thread context
2037 while (!list_empty(zombie_list)) {
2041 ni = list_entry(zombie_list->next,
2042 struct lnet_ni, ni_netlist);
2043 list_del_init(&ni->ni_netlist);
2044 /* the ni should be in deleting state. If it's not it's
2046 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2047 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2050 /* still busy, add it back to zombie list */
2051 list_add(&ni->ni_netlist, zombie_list);
2055 if (!list_empty(&ni->ni_netlist)) {
2056 lnet_net_unlock(LNET_LOCK_EX);
2058 if ((i & (-i)) == i) {
2060 "Waiting for zombie LNI %s\n",
2061 libcfs_nid2str(ni->ni_nid));
2063 set_current_state(TASK_UNINTERRUPTIBLE);
2064 schedule_timeout(cfs_time_seconds(1));
2065 lnet_net_lock(LNET_LOCK_EX);
2069 lnet_net_unlock(LNET_LOCK_EX);
2071 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2073 LASSERT(!in_interrupt());
2074 (net->net_lnd->lnd_shutdown)(ni);
2077 CDEBUG(D_LNI, "Removed LNI %s\n",
2078 libcfs_nid2str(ni->ni_nid));
2082 lnet_net_lock(LNET_LOCK_EX);
2086 /* shutdown down the NI and release refcount */
2088 lnet_shutdown_lndni(struct lnet_ni *ni)
2091 struct lnet_net *net = ni->ni_net;
2093 lnet_net_lock(LNET_LOCK_EX);
2095 ni->ni_state = LNET_NI_STATE_DELETING;
2097 lnet_ni_unlink_locked(ni);
2098 lnet_incr_dlc_seq();
2099 lnet_net_unlock(LNET_LOCK_EX);
2101 /* clear messages for this NI on the lazy portal */
2102 for (i = 0; i < the_lnet.ln_nportals; i++)
2103 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2105 lnet_net_lock(LNET_LOCK_EX);
2106 lnet_clear_zombies_nis_locked(net);
2107 lnet_net_unlock(LNET_LOCK_EX);
2111 lnet_shutdown_lndnet(struct lnet_net *net)
2115 lnet_net_lock(LNET_LOCK_EX);
2117 net->net_state = LNET_NET_STATE_DELETING;
2119 list_del_init(&net->net_list);
2121 while (!list_empty(&net->net_ni_list)) {
2122 ni = list_entry(net->net_ni_list.next,
2123 struct lnet_ni, ni_netlist);
2124 lnet_net_unlock(LNET_LOCK_EX);
2125 lnet_shutdown_lndni(ni);
2126 lnet_net_lock(LNET_LOCK_EX);
2129 lnet_net_unlock(LNET_LOCK_EX);
2131 /* Do peer table cleanup for this net */
2132 lnet_peer_tables_cleanup(net);
2134 lnet_net_lock(LNET_LOCK_EX);
2136 * decrement ref count on lnd only when the entire network goes
2139 net->net_lnd->lnd_refcount--;
2141 lnet_net_unlock(LNET_LOCK_EX);
2147 lnet_shutdown_lndnets(void)
2149 struct lnet_net *net;
2150 struct list_head resend;
2151 struct lnet_msg *msg, *tmp;
2153 INIT_LIST_HEAD(&resend);
2155 /* NB called holding the global mutex */
2157 /* All quiet on the API front */
2158 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2159 LASSERT(the_lnet.ln_refcount == 0);
2161 lnet_net_lock(LNET_LOCK_EX);
2162 the_lnet.ln_state = LNET_STATE_STOPPING;
2164 while (!list_empty(&the_lnet.ln_nets)) {
2166 * move the nets to the zombie list to avoid them being
2167 * picked up for new work. LONET is also included in the
2168 * Nets that will be moved to the zombie list
2170 net = list_entry(the_lnet.ln_nets.next,
2171 struct lnet_net, net_list);
2172 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2175 /* Drop the cached loopback Net. */
2176 if (the_lnet.ln_loni != NULL) {
2177 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2178 the_lnet.ln_loni = NULL;
2180 lnet_net_unlock(LNET_LOCK_EX);
2182 /* iterate through the net zombie list and delete each net */
2183 while (!list_empty(&the_lnet.ln_net_zombie)) {
2184 net = list_entry(the_lnet.ln_net_zombie.next,
2185 struct lnet_net, net_list);
2186 lnet_shutdown_lndnet(net);
2189 spin_lock(&the_lnet.ln_msg_resend_lock);
2190 list_splice(&the_lnet.ln_msg_resend, &resend);
2191 spin_unlock(&the_lnet.ln_msg_resend_lock);
2193 list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2194 list_del_init(&msg->msg_list);
2195 msg->msg_no_resend = true;
2196 lnet_finalize(msg, -ECANCELED);
2199 lnet_net_lock(LNET_LOCK_EX);
2200 the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2201 lnet_net_unlock(LNET_LOCK_EX);
2205 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2208 struct lnet_tx_queue *tq;
2210 struct lnet_net *net = ni->ni_net;
2212 mutex_lock(&the_lnet.ln_lnd_mutex);
2215 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2216 ni->ni_lnd_tunables_set = true;
2219 rc = (net->net_lnd->lnd_startup)(ni);
2221 mutex_unlock(&the_lnet.ln_lnd_mutex);
2224 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2225 rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2226 lnet_net_lock(LNET_LOCK_EX);
2227 net->net_lnd->lnd_refcount--;
2228 lnet_net_unlock(LNET_LOCK_EX);
2233 ni->ni_state = LNET_NI_STATE_ACTIVE;
2236 /* We keep a reference on the loopback net through the loopback NI */
2237 if (net->net_lnd->lnd_type == LOLND) {
2239 LASSERT(the_lnet.ln_loni == NULL);
2240 the_lnet.ln_loni = ni;
2241 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2242 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2243 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2244 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2248 if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2249 ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2250 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2251 libcfs_lnd2str(net->net_lnd->lnd_type),
2252 ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2254 /* shutdown the NI since if we get here then it must've already
2257 lnet_shutdown_lndni(ni);
2261 cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2262 tq->tq_credits_min =
2263 tq->tq_credits_max =
2264 tq->tq_credits = lnet_ni_tq_credits(ni);
2267 atomic_set(&ni->ni_tx_credits,
2268 lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2269 atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2271 CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2272 libcfs_nid2str(ni->ni_nid),
2273 ni->ni_net->net_tunables.lct_peer_tx_credits,
2274 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2275 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2276 ni->ni_net->net_tunables.lct_peer_timeout);
2285 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2288 struct lnet_net *net_l = NULL;
2289 struct list_head local_ni_list;
2293 struct lnet_lnd *lnd;
2295 net->net_tunables.lct_peer_timeout;
2297 net->net_tunables.lct_max_tx_credits;
2298 int peerrtrcredits =
2299 net->net_tunables.lct_peer_rtr_credits;
2301 INIT_LIST_HEAD(&local_ni_list);
2304 * make sure that this net is unique. If it isn't then
2305 * we are adding interfaces to an already existing network, and
2306 * 'net' is just a convenient way to pass in the list.
2307 * if it is unique we need to find the LND and load it if
2310 if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2311 lnd_type = LNET_NETTYP(net->net_id);
2313 mutex_lock(&the_lnet.ln_lnd_mutex);
2314 lnd = lnet_find_lnd_by_type(lnd_type);
2317 mutex_unlock(&the_lnet.ln_lnd_mutex);
2318 rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2319 mutex_lock(&the_lnet.ln_lnd_mutex);
2321 lnd = lnet_find_lnd_by_type(lnd_type);
2323 mutex_unlock(&the_lnet.ln_lnd_mutex);
2324 CERROR("Can't load LND %s, module %s, rc=%d\n",
2325 libcfs_lnd2str(lnd_type),
2326 libcfs_lnd2modname(lnd_type), rc);
2327 #ifndef HAVE_MODULE_LOADING_SUPPORT
2328 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2329 "compiled with kernel module "
2330 "loading support.");
2337 lnet_net_lock(LNET_LOCK_EX);
2338 lnd->lnd_refcount++;
2339 lnet_net_unlock(LNET_LOCK_EX);
2343 mutex_unlock(&the_lnet.ln_lnd_mutex);
2349 * net_l: if the network being added is unique then net_l
2350 * will point to that network
2351 * if the network being added is not unique then
2352 * net_l points to the existing network.
2354 * When we enter the loop below, we'll pick NIs off he
2355 * network beign added and start them up, then add them to
2356 * a local ni list. Once we've successfully started all
2357 * the NIs then we join the local NI list (of started up
2358 * networks) with the net_l->net_ni_list, which should
2359 * point to the correct network to add the new ni list to
2361 * If any of the new NIs fail to start up, then we want to
2362 * iterate through the local ni list, which should include
2363 * any NIs which were successfully started up, and shut
2366 * After than we want to delete the network being added,
2367 * to avoid a memory leak.
2371 * When a network uses TCP bonding then all its interfaces
2372 * must be specified when the network is first defined: the
2373 * TCP bonding code doesn't allow for interfaces to be added
2376 if (net_l != net && net_l != NULL && use_tcp_bonding &&
2377 LNET_NETTYP(net_l->net_id) == SOCKLND) {
2382 while (!list_empty(&net->net_ni_added)) {
2383 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2385 list_del_init(&ni->ni_netlist);
2387 /* make sure that the the NI we're about to start
2388 * up is actually unique. if it's not fail. */
2389 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2390 ni->ni_interfaces[0])) {
2395 /* adjust the pointer the parent network, just in case it
2396 * the net is a duplicate */
2399 rc = lnet_startup_lndni(ni, tun);
2401 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2402 ni->ni_net->net_lnd->lnd_query != NULL);
2408 list_add_tail(&ni->ni_netlist, &local_ni_list);
2413 lnet_net_lock(LNET_LOCK_EX);
2414 list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2415 lnet_incr_dlc_seq();
2416 lnet_net_unlock(LNET_LOCK_EX);
2418 /* if the network is not unique then we don't want to keep
2419 * it around after we're done. Free it. Otherwise add that
2420 * net to the global the_lnet.ln_nets */
2421 if (net_l != net && net_l != NULL) {
2423 * TODO - note. currently the tunables can not be updated
2428 net->net_state = LNET_NET_STATE_ACTIVE;
2430 * restore tunables after it has been overwitten by the
2433 if (peer_timeout != -1)
2434 net->net_tunables.lct_peer_timeout = peer_timeout;
2435 if (maxtxcredits != -1)
2436 net->net_tunables.lct_max_tx_credits = maxtxcredits;
2437 if (peerrtrcredits != -1)
2438 net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2440 lnet_net_lock(LNET_LOCK_EX);
2441 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2442 lnet_net_unlock(LNET_LOCK_EX);
2445 /* update net count */
2446 lnet_current_net_count = lnet_get_net_count();
2452 * shutdown the new NIs that are being started up
2453 * free the NET being started
2455 while (!list_empty(&local_ni_list)) {
2456 ni = list_entry(local_ni_list.next, struct lnet_ni,
2459 lnet_shutdown_lndni(ni);
2469 lnet_startup_lndnets(struct list_head *netlist)
2471 struct lnet_net *net;
2476 * Change to running state before bringing up the LNDs. This
2477 * allows lnet_shutdown_lndnets() to assert that we've passed
2480 lnet_net_lock(LNET_LOCK_EX);
2481 the_lnet.ln_state = LNET_STATE_RUNNING;
2482 lnet_net_unlock(LNET_LOCK_EX);
2484 while (!list_empty(netlist)) {
2485 net = list_entry(netlist->next, struct lnet_net, net_list);
2486 list_del_init(&net->net_list);
2488 rc = lnet_startup_lndnet(net, NULL);
2498 lnet_shutdown_lndnets();
2504 * Initialize LNet library.
2506 * Automatically called at module loading time. Caller has to call
2507 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2508 * latter returned 0. It must be called exactly once.
2510 * \retval 0 on success
2511 * \retval -ve on failures.
2513 int lnet_lib_init(void)
2517 lnet_assert_wire_constants();
2519 /* refer to global cfs_cpt_table for now */
2520 the_lnet.ln_cpt_table = cfs_cpt_table;
2521 the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table);
2523 LASSERT(the_lnet.ln_cpt_number > 0);
2524 if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2525 /* we are under risk of consuming all lh_cookie */
2526 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2527 "please change setting of CPT-table and retry\n",
2528 the_lnet.ln_cpt_number, LNET_CPT_MAX);
2532 while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2533 the_lnet.ln_cpt_bits++;
2535 rc = lnet_create_locks();
2537 CERROR("Can't create LNet global locks: %d\n", rc);
2541 the_lnet.ln_refcount = 0;
2542 INIT_LIST_HEAD(&the_lnet.ln_lnds);
2543 INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2544 INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2546 /* The hash table size is the number of bits it takes to express the set
2547 * ln_num_routes, minus 1 (better to under estimate than over so we
2548 * don't waste memory). */
2549 if (rnet_htable_size <= 0)
2550 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2551 else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2552 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2553 the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2554 order_base_2(rnet_htable_size) - 1);
2556 /* All LNDs apart from the LOLND are in separate modules. They
2557 * register themselves when their module loads, and unregister
2558 * themselves when their module is unloaded. */
2559 lnet_register_lnd(&the_lolnd);
2564 * Finalize LNet library.
2566 * \pre lnet_lib_init() called with success.
2567 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2569 void lnet_lib_exit(void)
2571 LASSERT(the_lnet.ln_refcount == 0);
2573 while (!list_empty(&the_lnet.ln_lnds))
2574 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2575 struct lnet_lnd, lnd_list));
2576 lnet_destroy_locks();
2580 * Set LNet PID and start LNet interfaces, routing, and forwarding.
2582 * Users must call this function at least once before any other functions.
2583 * For each successful call there must be a corresponding call to
2584 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2587 * The PID used by LNet may be different from the one requested.
2590 * \param requested_pid PID requested by the caller.
2592 * \return >= 0 on success, and < 0 error code on failures.
2595 LNetNIInit(lnet_pid_t requested_pid)
2597 int im_a_router = 0;
2600 struct lnet_ping_buffer *pbuf;
2601 struct lnet_handle_md ping_mdh;
2602 struct list_head net_head;
2603 struct lnet_net *net;
2605 INIT_LIST_HEAD(&net_head);
2607 mutex_lock(&the_lnet.ln_api_mutex);
2609 CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2611 if (the_lnet.ln_refcount > 0) {
2612 rc = the_lnet.ln_refcount++;
2613 mutex_unlock(&the_lnet.ln_api_mutex);
2617 rc = lnet_prepare(requested_pid);
2619 mutex_unlock(&the_lnet.ln_api_mutex);
2623 /* create a network for Loopback network */
2624 net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2627 goto err_empty_list;
2630 /* Add in the loopback NI */
2631 if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2633 goto err_empty_list;
2636 /* If LNet is being initialized via DLC it is possible
2637 * that the user requests not to load module parameters (ones which
2638 * are supported by DLC) on initialization. Therefore, make sure not
2639 * to load networks, routes and forwarding from module parameters
2640 * in this case. On cleanup in case of failure only clean up
2641 * routes if it has been loaded */
2642 if (!the_lnet.ln_nis_from_mod_params) {
2643 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2646 goto err_empty_list;
2649 ni_count = lnet_startup_lndnets(&net_head);
2652 goto err_empty_list;
2655 if (!the_lnet.ln_nis_from_mod_params) {
2656 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2658 goto err_shutdown_lndnis;
2660 rc = lnet_rtrpools_alloc(im_a_router);
2662 goto err_destroy_routes;
2665 rc = lnet_acceptor_start();
2667 goto err_destroy_routes;
2669 the_lnet.ln_refcount = 1;
2670 /* Now I may use my own API functions... */
2672 rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2674 goto err_acceptor_stop;
2676 lnet_ping_target_update(pbuf, ping_mdh);
2678 rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2680 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2684 rc = lnet_push_target_init();
2688 rc = lnet_peer_discovery_start();
2690 goto err_destroy_push_target;
2692 rc = lnet_monitor_thr_start();
2694 goto err_stop_discovery_thr;
2697 lnet_router_debugfs_init();
2699 mutex_unlock(&the_lnet.ln_api_mutex);
2701 complete_all(&the_lnet.ln_started);
2703 /* wait for all routers to start */
2704 lnet_wait_router_start();
2708 err_stop_discovery_thr:
2709 lnet_peer_discovery_stop();
2710 err_destroy_push_target:
2711 lnet_push_target_fini();
2713 lnet_ping_target_fini();
2715 the_lnet.ln_refcount = 0;
2716 lnet_acceptor_stop();
2718 if (!the_lnet.ln_nis_from_mod_params)
2719 lnet_destroy_routes();
2720 err_shutdown_lndnis:
2721 lnet_shutdown_lndnets();
2725 mutex_unlock(&the_lnet.ln_api_mutex);
2726 while (!list_empty(&net_head)) {
2727 struct lnet_net *net;
2729 net = list_entry(net_head.next, struct lnet_net, net_list);
2730 list_del_init(&net->net_list);
2735 EXPORT_SYMBOL(LNetNIInit);
2738 * Stop LNet interfaces, routing, and forwarding.
2740 * Users must call this function once for each successful call to LNetNIInit().
2741 * Once the LNetNIFini() operation has been started, the results of pending
2742 * API operations are undefined.
2744 * \return always 0 for current implementation.
2749 mutex_lock(&the_lnet.ln_api_mutex);
2751 LASSERT(the_lnet.ln_refcount > 0);
2753 if (the_lnet.ln_refcount != 1) {
2754 the_lnet.ln_refcount--;
2756 LASSERT(!the_lnet.ln_niinit_self);
2760 lnet_router_debugfs_fini();
2761 lnet_monitor_thr_stop();
2762 lnet_peer_discovery_stop();
2763 lnet_push_target_fini();
2764 lnet_ping_target_fini();
2766 /* Teardown fns that use my own API functions BEFORE here */
2767 the_lnet.ln_refcount = 0;
2769 lnet_acceptor_stop();
2770 lnet_destroy_routes();
2771 lnet_shutdown_lndnets();
2775 mutex_unlock(&the_lnet.ln_api_mutex);
2778 EXPORT_SYMBOL(LNetNIFini);
2781 * Grabs the ni data from the ni structure and fills the out
2784 * \param[in] ni network interface structure
2785 * \param[out] cfg_ni NI config information
2786 * \param[out] tun network and LND tunables
2789 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2790 struct lnet_ioctl_config_lnd_tunables *tun,
2791 struct lnet_ioctl_element_stats *stats,
2794 size_t min_size = 0;
2797 if (!ni || !cfg_ni || !tun)
2800 if (ni->ni_interfaces[0] != NULL) {
2801 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2802 if (ni->ni_interfaces[i] != NULL) {
2803 strncpy(cfg_ni->lic_ni_intf[i],
2804 ni->ni_interfaces[i],
2805 sizeof(cfg_ni->lic_ni_intf[i]));
2810 cfg_ni->lic_nid = ni->ni_nid;
2811 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2812 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2814 cfg_ni->lic_status = ni->ni_status->ns_status;
2815 cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2816 cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2818 memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2821 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2822 LNET_STATS_TYPE_SEND);
2823 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2824 LNET_STATS_TYPE_RECV);
2825 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2826 LNET_STATS_TYPE_DROP);
2830 * tun->lt_tun will always be present, but in order to be
2831 * backwards compatible, we need to deal with the cases when
2832 * tun->lt_tun is smaller than what the kernel has, because it
2833 * comes from an older version of a userspace program, then we'll
2834 * need to copy as much information as we have available space.
2836 min_size = tun_size - sizeof(tun->lt_cmn);
2837 memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2839 /* copy over the cpts */
2840 if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2841 ni->ni_cpts == NULL) {
2842 for (i = 0; i < ni->ni_ncpts; i++)
2843 cfg_ni->lic_cpts[i] = i;
2846 ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2847 i < LNET_MAX_SHOW_NUM_CPT;
2849 cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2851 cfg_ni->lic_ncpts = ni->ni_ncpts;
2855 * NOTE: This is a legacy function left in the code to be backwards
2856 * compatible with older userspace programs. It should eventually be
2859 * Grabs the ni data from the ni structure and fills the out
2862 * \param[in] ni network interface structure
2863 * \param[out] config config information
2866 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2867 struct lnet_ioctl_config_data *config)
2869 struct lnet_ioctl_net_config *net_config;
2870 struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2871 size_t min_size, tunable_size = 0;
2877 net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2881 BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2882 ARRAY_SIZE(net_config->ni_interfaces));
2884 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2885 if (!ni->ni_interfaces[i])
2888 strncpy(net_config->ni_interfaces[i],
2889 ni->ni_interfaces[i],
2890 sizeof(net_config->ni_interfaces[i]));
2893 config->cfg_nid = ni->ni_nid;
2894 config->cfg_config_u.cfg_net.net_peer_timeout =
2895 ni->ni_net->net_tunables.lct_peer_timeout;
2896 config->cfg_config_u.cfg_net.net_max_tx_credits =
2897 ni->ni_net->net_tunables.lct_max_tx_credits;
2898 config->cfg_config_u.cfg_net.net_peer_tx_credits =
2899 ni->ni_net->net_tunables.lct_peer_tx_credits;
2900 config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2901 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2903 if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2904 net_config->ni_status = LNET_NI_STATUS_UP;
2906 net_config->ni_status = ni->ni_status->ns_status;
2909 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2911 for (i = 0; i < num_cpts; i++)
2912 net_config->ni_cpts[i] = ni->ni_cpts[i];
2914 config->cfg_ncpts = num_cpts;
2918 * See if user land tools sent in a newer and larger version
2919 * of struct lnet_tunables than what the kernel uses.
2921 min_size = sizeof(*config) + sizeof(*net_config);
2923 if (config->cfg_hdr.ioc_len > min_size)
2924 tunable_size = config->cfg_hdr.ioc_len - min_size;
2926 /* Don't copy too much data to user space */
2927 min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2928 lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2930 if (lnd_cfg && min_size) {
2931 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2932 config->cfg_config_u.cfg_net.net_interface_count = 1;
2934 /* Tell user land that kernel side has less data */
2935 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2936 min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2937 config->cfg_hdr.ioc_len -= min_size;
2943 lnet_get_ni_idx_locked(int idx)
2946 struct lnet_net *net;
2948 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2949 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2959 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2962 struct lnet_net *net = mynet;
2965 * It is possible that the net has been cleaned out while there is
2966 * a message being sent. This function accessed the net without
2967 * checking if the list is empty
2971 net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2973 if (list_empty(&net->net_ni_list))
2975 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2981 if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2982 /* if you reached the end of the ni list and the net is
2983 * specified, then there are no more nis in that net */
2987 /* we reached the end of this net ni list. move to the
2989 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2990 /* no more nets and no more NIs. */
2993 /* get the next net */
2994 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2996 if (list_empty(&net->net_ni_list))
2998 /* get the ni on it */
2999 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3005 if (list_empty(&prev->ni_netlist))
3008 /* there are more nis left */
3009 ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3015 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3020 int idx = config->cfg_count;
3022 cpt = lnet_net_lock_current();
3024 ni = lnet_get_ni_idx_locked(idx);
3029 lnet_fill_ni_info_legacy(ni, config);
3033 lnet_net_unlock(cpt);
3038 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3039 struct lnet_ioctl_config_lnd_tunables *tun,
3040 struct lnet_ioctl_element_stats *stats,
3047 if (!cfg_ni || !tun || !stats)
3050 cpt = lnet_net_lock_current();
3052 ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3057 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3061 lnet_net_unlock(cpt);
3065 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3074 cpt = lnet_net_lock_current();
3076 ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3079 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3083 lnet_net_unlock(cpt);
3088 static int lnet_add_net_common(struct lnet_net *net,
3089 struct lnet_ioctl_config_lnd_tunables *tun)
3092 struct lnet_ping_buffer *pbuf;
3093 struct lnet_handle_md ping_mdh;
3095 struct lnet_remotenet *rnet;
3097 int num_acceptor_nets;
3099 lnet_net_lock(LNET_LOCK_EX);
3100 rnet = lnet_find_rnet_locked(net->net_id);
3101 lnet_net_unlock(LNET_LOCK_EX);
3103 * make sure that the net added doesn't invalidate the current
3104 * configuration LNet is keeping
3107 CERROR("Adding net %s will invalidate routing configuration\n",
3108 libcfs_net2str(net->net_id));
3114 * make sure you calculate the correct number of slots in the ping
3115 * buffer. Since the ping info is a flattened list of all the NIs,
3116 * we should allocate enough slots to accomodate the number of NIs
3117 * which will be added.
3119 * since ni hasn't been configured yet, use
3120 * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3122 net_ni_count = lnet_get_net_ni_count_pre(net);
3124 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3125 net_ni_count + lnet_get_ni_count(),
3133 memcpy(&net->net_tunables,
3134 &tun->lt_cmn, sizeof(net->net_tunables));
3136 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3139 * before starting this network get a count of the current TCP
3140 * networks which require the acceptor thread running. If that
3141 * count is == 0 before we start up this network, then we'd want to
3142 * start up the acceptor thread after starting up this network
3144 num_acceptor_nets = lnet_count_acceptor_nets();
3146 net_id = net->net_id;
3148 rc = lnet_startup_lndnet(net,
3149 (tun) ? &tun->lt_tun : NULL);
3153 lnet_net_lock(LNET_LOCK_EX);
3154 net = lnet_get_net_locked(net_id);
3155 lnet_net_unlock(LNET_LOCK_EX);
3160 * Start the acceptor thread if this is the first network
3161 * being added that requires the thread.
3163 if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3164 rc = lnet_acceptor_start();
3166 /* shutdown the net that we just started */
3167 CERROR("Failed to start up acceptor thread\n");
3168 lnet_shutdown_lndnet(net);
3173 lnet_net_lock(LNET_LOCK_EX);
3174 lnet_peer_net_added(net);
3175 lnet_net_unlock(LNET_LOCK_EX);
3177 lnet_ping_target_update(pbuf, ping_mdh);
3182 lnet_ping_md_unlink(pbuf, &ping_mdh);
3183 lnet_ping_buffer_decref(pbuf);
3187 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3188 struct lnet_ioctl_config_lnd_tunables *tun)
3190 struct lnet_net *net;
3193 struct list_head net_head;
3195 INIT_LIST_HEAD(&net_head);
3197 rc = lnet_parse_ip2nets(&nets, ip2nets);
3201 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3205 mutex_lock(&the_lnet.ln_api_mutex);
3206 while (!list_empty(&net_head)) {
3207 net = list_entry(net_head.next, struct lnet_net, net_list);
3208 list_del_init(&net->net_list);
3209 rc = lnet_add_net_common(net, tun);
3215 mutex_unlock(&the_lnet.ln_api_mutex);
3217 while (!list_empty(&net_head)) {
3218 net = list_entry(net_head.next, struct lnet_net, net_list);
3219 list_del_init(&net->net_list);
3225 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3227 struct lnet_net *net;
3229 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3231 __u32 net_id, lnd_type;
3233 /* get the tunables if they are available */
3234 if (conf->lic_cfg_hdr.ioc_len >=
3235 sizeof(*conf) + sizeof(*tun))
3236 tun = (struct lnet_ioctl_config_lnd_tunables *)
3239 /* handle legacy ip2nets from DLC */
3240 if (conf->lic_legacy_ip2nets[0] != '\0')
3241 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3244 net_id = LNET_NIDNET(conf->lic_nid);
3245 lnd_type = LNET_NETTYP(net_id);
3247 if (!libcfs_isknown_lnd(lnd_type)) {
3248 CERROR("No valid net and lnd information provided\n");
3252 net = lnet_net_alloc(net_id, NULL);
3256 for (i = 0; i < conf->lic_ncpts; i++) {
3257 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3261 ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3262 conf->lic_ni_intf[0]);
3266 mutex_lock(&the_lnet.ln_api_mutex);
3268 rc = lnet_add_net_common(net, tun);
3270 mutex_unlock(&the_lnet.ln_api_mutex);
3275 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3277 struct lnet_net *net;
3279 __u32 net_id = LNET_NIDNET(conf->lic_nid);
3280 struct lnet_ping_buffer *pbuf;
3281 struct lnet_handle_md ping_mdh;
3286 /* don't allow userspace to shutdown the LOLND */
3287 if (LNET_NETTYP(net_id) == LOLND)
3290 mutex_lock(&the_lnet.ln_api_mutex);
3294 net = lnet_get_net_locked(net_id);
3296 CERROR("net %s not found\n",
3297 libcfs_net2str(net_id));
3302 addr = LNET_NIDADDR(conf->lic_nid);
3304 /* remove the entire net */
3305 net_count = lnet_get_net_ni_count_locked(net);
3309 /* create and link a new ping info, before removing the old one */
3310 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3311 lnet_get_ni_count() - net_count,
3314 goto unlock_api_mutex;
3316 lnet_shutdown_lndnet(net);
3318 if (lnet_count_acceptor_nets() == 0)
3319 lnet_acceptor_stop();
3321 lnet_ping_target_update(pbuf, ping_mdh);
3323 goto unlock_api_mutex;
3326 ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3328 CERROR("nid %s not found\n",
3329 libcfs_nid2str(conf->lic_nid));
3334 net_count = lnet_get_net_ni_count_locked(net);
3338 /* create and link a new ping info, before removing the old one */
3339 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3340 lnet_get_ni_count() - 1, false);
3342 goto unlock_api_mutex;
3344 lnet_shutdown_lndni(ni);
3346 if (lnet_count_acceptor_nets() == 0)
3347 lnet_acceptor_stop();
3349 lnet_ping_target_update(pbuf, ping_mdh);
3351 /* check if the net is empty and remove it if it is */
3353 lnet_shutdown_lndnet(net);
3355 goto unlock_api_mutex;
3360 mutex_unlock(&the_lnet.ln_api_mutex);
3366 * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3367 * They are only expected to be called for unique networks.
3368 * That can be as a result of older DLC library
3369 * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3372 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3374 struct lnet_net *net;
3375 struct list_head net_head;
3377 struct lnet_ioctl_config_lnd_tunables tun;
3378 char *nets = conf->cfg_config_u.cfg_net.net_intf;
3380 INIT_LIST_HEAD(&net_head);
3382 /* Create a net/ni structures for the network string */
3383 rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3385 return rc == 0 ? -EINVAL : rc;
3387 mutex_lock(&the_lnet.ln_api_mutex);
3390 rc = -EINVAL; /* only add one network per call */
3391 goto out_unlock_clean;
3394 net = list_entry(net_head.next, struct lnet_net, net_list);
3395 list_del_init(&net->net_list);
3397 LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3399 memset(&tun, 0, sizeof(tun));
3401 tun.lt_cmn.lct_peer_timeout =
3402 conf->cfg_config_u.cfg_net.net_peer_timeout;
3403 tun.lt_cmn.lct_peer_tx_credits =
3404 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3405 tun.lt_cmn.lct_peer_rtr_credits =
3406 conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3407 tun.lt_cmn.lct_max_tx_credits =
3408 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3410 rc = lnet_add_net_common(net, &tun);
3413 mutex_unlock(&the_lnet.ln_api_mutex);
3414 while (!list_empty(&net_head)) {
3415 /* net_head list is empty in success case */
3416 net = list_entry(net_head.next, struct lnet_net, net_list);
3417 list_del_init(&net->net_list);
3424 lnet_dyn_del_net(__u32 net_id)
3426 struct lnet_net *net;
3427 struct lnet_ping_buffer *pbuf;
3428 struct lnet_handle_md ping_mdh;
3432 /* don't allow userspace to shutdown the LOLND */
3433 if (LNET_NETTYP(net_id) == LOLND)
3436 mutex_lock(&the_lnet.ln_api_mutex);
3440 net = lnet_get_net_locked(net_id);
3447 net_ni_count = lnet_get_net_ni_count_locked(net);
3451 /* create and link a new ping info, before removing the old one */
3452 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3453 lnet_get_ni_count() - net_ni_count, false);
3457 lnet_shutdown_lndnet(net);
3459 if (lnet_count_acceptor_nets() == 0)
3460 lnet_acceptor_stop();
3462 lnet_ping_target_update(pbuf, ping_mdh);
3465 mutex_unlock(&the_lnet.ln_api_mutex);
3470 void lnet_incr_dlc_seq(void)
3472 atomic_inc(&lnet_dlc_seq_no);
3475 __u32 lnet_get_dlc_seq_locked(void)
3477 return atomic_read(&lnet_dlc_seq_no);
3481 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3483 struct lnet_net *net;
3486 lnet_net_lock(LNET_LOCK_EX);
3487 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3488 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3489 if (ni->ni_nid == nid || all) {
3490 atomic_set(&ni->ni_healthv, value);
3491 if (list_empty(&ni->ni_recovery) &&
3492 value < LNET_MAX_HEALTH_VALUE) {
3493 CERROR("manually adding local NI %s to recovery\n",
3494 libcfs_nid2str(ni->ni_nid));
3495 list_add_tail(&ni->ni_recovery,
3496 &the_lnet.ln_mt_localNIRecovq);
3497 lnet_ni_addref_locked(ni, 0);
3500 lnet_net_unlock(LNET_LOCK_EX);
3506 lnet_net_unlock(LNET_LOCK_EX);
3510 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3514 lnet_nid_t nid = stats->hlni_nid;
3516 cpt = lnet_net_lock_current();
3517 ni = lnet_nid2ni_locked(nid, cpt);
3524 stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3525 stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3526 stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3527 stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3528 stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3529 stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3530 stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3533 lnet_net_unlock(cpt);
3539 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3544 lnet_net_lock(LNET_LOCK_EX);
3545 list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3546 list->rlst_nid_array[i] = ni->ni_nid;
3548 if (i >= LNET_MAX_SHOW_NUM_NID)
3551 lnet_net_unlock(LNET_LOCK_EX);
3552 list->rlst_num_nids = i;
3558 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3560 struct lnet_peer_ni *lpni;
3563 lnet_net_lock(LNET_LOCK_EX);
3564 list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3565 list->rlst_nid_array[i] = lpni->lpni_nid;
3567 if (i >= LNET_MAX_SHOW_NUM_NID)
3570 lnet_net_unlock(LNET_LOCK_EX);
3571 list->rlst_num_nids = i;
3577 * LNet ioctl handler.
3581 LNetCtl(unsigned int cmd, void *arg)
3583 struct libcfs_ioctl_data *data = arg;
3584 struct lnet_ioctl_config_data *config;
3585 struct lnet_process_id id = {0};
3589 BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3590 sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3593 case IOC_LIBCFS_GET_NI:
3594 rc = LNetGetId(data->ioc_count, &id);
3595 data->ioc_nid = id.nid;
3598 case IOC_LIBCFS_FAIL_NID:
3599 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3601 case IOC_LIBCFS_ADD_ROUTE: {
3602 /* default router sensitivity to 1 */
3603 unsigned int sensitivity = 1;
3606 if (config->cfg_hdr.ioc_len < sizeof(*config))
3609 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3611 config->cfg_config_u.cfg_route.rtr_sensitivity;
3614 mutex_lock(&the_lnet.ln_api_mutex);
3615 rc = lnet_add_route(config->cfg_net,
3616 config->cfg_config_u.cfg_route.rtr_hop,
3618 config->cfg_config_u.cfg_route.
3619 rtr_priority, sensitivity);
3620 mutex_unlock(&the_lnet.ln_api_mutex);
3624 case IOC_LIBCFS_DEL_ROUTE:
3627 if (config->cfg_hdr.ioc_len < sizeof(*config))
3630 mutex_lock(&the_lnet.ln_api_mutex);
3631 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3632 mutex_unlock(&the_lnet.ln_api_mutex);
3635 case IOC_LIBCFS_GET_ROUTE:
3638 if (config->cfg_hdr.ioc_len < sizeof(*config))
3641 mutex_lock(&the_lnet.ln_api_mutex);
3642 rc = lnet_get_route(config->cfg_count,
3644 &config->cfg_config_u.cfg_route.rtr_hop,
3646 &config->cfg_config_u.cfg_route.rtr_flags,
3647 &config->cfg_config_u.cfg_route.
3649 &config->cfg_config_u.cfg_route.
3651 mutex_unlock(&the_lnet.ln_api_mutex);
3654 case IOC_LIBCFS_GET_LOCAL_NI: {
3655 struct lnet_ioctl_config_ni *cfg_ni;
3656 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3657 struct lnet_ioctl_element_stats *stats;
3662 /* get the tunables if they are available */
3663 if (cfg_ni->lic_cfg_hdr.ioc_len <
3664 sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3667 stats = (struct lnet_ioctl_element_stats *)
3669 tun = (struct lnet_ioctl_config_lnd_tunables *)
3670 (cfg_ni->lic_bulk + sizeof(*stats));
3672 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3675 mutex_lock(&the_lnet.ln_api_mutex);
3676 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3677 mutex_unlock(&the_lnet.ln_api_mutex);
3681 case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3682 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3684 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3687 mutex_lock(&the_lnet.ln_api_mutex);
3688 rc = lnet_get_ni_stats(msg_stats);
3689 mutex_unlock(&the_lnet.ln_api_mutex);
3694 case IOC_LIBCFS_GET_NET: {
3695 size_t total = sizeof(*config) +
3696 sizeof(struct lnet_ioctl_net_config);
3699 if (config->cfg_hdr.ioc_len < total)
3702 mutex_lock(&the_lnet.ln_api_mutex);
3703 rc = lnet_get_net_config(config);
3704 mutex_unlock(&the_lnet.ln_api_mutex);
3708 case IOC_LIBCFS_GET_LNET_STATS:
3710 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3712 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3715 mutex_lock(&the_lnet.ln_api_mutex);
3716 lnet_counters_get(&lnet_stats->st_cntrs);
3717 mutex_unlock(&the_lnet.ln_api_mutex);
3721 case IOC_LIBCFS_CONFIG_RTR:
3724 if (config->cfg_hdr.ioc_len < sizeof(*config))
3727 mutex_lock(&the_lnet.ln_api_mutex);
3728 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3729 rc = lnet_rtrpools_enable();
3730 mutex_unlock(&the_lnet.ln_api_mutex);
3733 lnet_rtrpools_disable();
3734 mutex_unlock(&the_lnet.ln_api_mutex);
3737 case IOC_LIBCFS_ADD_BUF:
3740 if (config->cfg_hdr.ioc_len < sizeof(*config))
3743 mutex_lock(&the_lnet.ln_api_mutex);
3744 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3746 config->cfg_config_u.cfg_buffers.
3748 config->cfg_config_u.cfg_buffers.
3750 mutex_unlock(&the_lnet.ln_api_mutex);
3753 case IOC_LIBCFS_SET_NUMA_RANGE: {
3754 struct lnet_ioctl_set_value *numa;
3756 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3758 lnet_net_lock(LNET_LOCK_EX);
3759 lnet_numa_range = numa->sv_value;
3760 lnet_net_unlock(LNET_LOCK_EX);
3764 case IOC_LIBCFS_GET_NUMA_RANGE: {
3765 struct lnet_ioctl_set_value *numa;
3767 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3769 numa->sv_value = lnet_numa_range;
3773 case IOC_LIBCFS_GET_BUF: {
3774 struct lnet_ioctl_pool_cfg *pool_cfg;
3775 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3779 if (config->cfg_hdr.ioc_len < total)
3782 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3784 mutex_lock(&the_lnet.ln_api_mutex);
3785 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3786 mutex_unlock(&the_lnet.ln_api_mutex);
3790 case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3791 struct lnet_ioctl_local_ni_hstats *stats = arg;
3793 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3796 mutex_lock(&the_lnet.ln_api_mutex);
3797 rc = lnet_get_local_ni_hstats(stats);
3798 mutex_unlock(&the_lnet.ln_api_mutex);
3803 case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3804 struct lnet_ioctl_recovery_list *list = arg;
3805 if (list->rlst_hdr.ioc_len < sizeof(*list))
3808 mutex_lock(&the_lnet.ln_api_mutex);
3809 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3810 rc = lnet_get_local_ni_recovery_list(list);
3812 rc = lnet_get_peer_ni_recovery_list(list);
3813 mutex_unlock(&the_lnet.ln_api_mutex);
3817 case IOC_LIBCFS_ADD_PEER_NI: {
3818 struct lnet_ioctl_peer_cfg *cfg = arg;
3820 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3823 mutex_lock(&the_lnet.ln_api_mutex);
3824 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3827 mutex_unlock(&the_lnet.ln_api_mutex);
3831 case IOC_LIBCFS_DEL_PEER_NI: {
3832 struct lnet_ioctl_peer_cfg *cfg = arg;
3834 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3837 mutex_lock(&the_lnet.ln_api_mutex);
3838 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3839 cfg->prcfg_cfg_nid);
3840 mutex_unlock(&the_lnet.ln_api_mutex);
3844 case IOC_LIBCFS_GET_PEER_INFO: {
3845 struct lnet_ioctl_peer *peer_info = arg;
3847 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3850 mutex_lock(&the_lnet.ln_api_mutex);
3851 rc = lnet_get_peer_ni_info(
3852 peer_info->pr_count,
3854 peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3855 &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3856 &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3857 &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3858 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3859 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3860 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3861 &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3862 mutex_unlock(&the_lnet.ln_api_mutex);
3866 case IOC_LIBCFS_GET_PEER_NI: {
3867 struct lnet_ioctl_peer_cfg *cfg = arg;
3869 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3872 mutex_lock(&the_lnet.ln_api_mutex);
3873 rc = lnet_get_peer_info(cfg,
3874 (void __user *)cfg->prcfg_bulk);
3875 mutex_unlock(&the_lnet.ln_api_mutex);
3879 case IOC_LIBCFS_GET_PEER_LIST: {
3880 struct lnet_ioctl_peer_cfg *cfg = arg;
3882 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3885 mutex_lock(&the_lnet.ln_api_mutex);
3886 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3887 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3888 mutex_unlock(&the_lnet.ln_api_mutex);
3892 case IOC_LIBCFS_SET_HEALHV: {
3893 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3895 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3897 if (cfg->rh_value < 0 ||
3898 cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3899 value = LNET_MAX_HEALTH_VALUE;
3901 value = cfg->rh_value;
3902 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3903 value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3904 "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3905 mutex_lock(&the_lnet.ln_api_mutex);
3906 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3907 lnet_ni_set_healthv(cfg->rh_nid, value,
3910 lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3912 mutex_unlock(&the_lnet.ln_api_mutex);
3916 case IOC_LIBCFS_NOTIFY_ROUTER: {
3917 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3919 /* The deadline passed in by the user should be some time in
3920 * seconds in the future since the UNIX epoch. We have to map
3921 * that deadline to the wall clock.
3923 deadline += ktime_get_seconds();
3924 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3928 case IOC_LIBCFS_LNET_DIST:
3929 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3930 if (rc < 0 && rc != -EHOSTUNREACH)
3933 data->ioc_u32[0] = rc;
3936 case IOC_LIBCFS_TESTPROTOCOMPAT:
3937 lnet_net_lock(LNET_LOCK_EX);
3938 the_lnet.ln_testprotocompat = data->ioc_flags;
3939 lnet_net_unlock(LNET_LOCK_EX);
3942 case IOC_LIBCFS_LNET_FAULT:
3943 return lnet_fault_ctl(data->ioc_flags, data);
3945 case IOC_LIBCFS_PING: {
3946 signed long timeout;
3948 id.nid = data->ioc_nid;
3949 id.pid = data->ioc_u32[0];
3951 /* If timeout is negative then set default of 3 minutes */
3952 if (((s32)data->ioc_u32[1] <= 0) ||
3953 data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3954 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3956 timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3958 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3959 data->ioc_plen1 / sizeof(struct lnet_process_id));
3964 data->ioc_count = rc;
3968 case IOC_LIBCFS_PING_PEER: {
3969 struct lnet_ioctl_ping_data *ping = arg;
3970 struct lnet_peer *lp;
3971 signed long timeout;
3973 /* If timeout is negative then set default of 3 minutes */
3974 if (((s32)ping->op_param) <= 0 ||
3975 ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3976 timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3978 timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3980 rc = lnet_ping(ping->ping_id, timeout,
3986 mutex_lock(&the_lnet.ln_api_mutex);
3987 lp = lnet_find_peer(ping->ping_id.nid);
3989 ping->ping_id.nid = lp->lp_primary_nid;
3990 ping->mr_info = lnet_peer_is_multi_rail(lp);
3991 lnet_peer_decref_locked(lp);
3993 mutex_unlock(&the_lnet.ln_api_mutex);
3995 ping->ping_count = rc;
3999 case IOC_LIBCFS_DISCOVER: {
4000 struct lnet_ioctl_ping_data *discover = arg;
4001 struct lnet_peer *lp;
4003 rc = lnet_discover(discover->ping_id, discover->op_param,
4005 discover->ping_count);
4009 mutex_lock(&the_lnet.ln_api_mutex);
4010 lp = lnet_find_peer(discover->ping_id.nid);
4012 discover->ping_id.nid = lp->lp_primary_nid;
4013 discover->mr_info = lnet_peer_is_multi_rail(lp);
4014 lnet_peer_decref_locked(lp);
4016 mutex_unlock(&the_lnet.ln_api_mutex);
4018 discover->ping_count = rc;
4023 ni = lnet_net2ni_addref(data->ioc_net);
4027 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4030 rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4037 EXPORT_SYMBOL(LNetCtl);
4039 void LNetDebugPeer(struct lnet_process_id id)
4041 lnet_debug_peer(id.nid);
4043 EXPORT_SYMBOL(LNetDebugPeer);
4046 * Determine if the specified peer \a nid is on the local node.
4048 * \param nid peer nid to check
4050 * \retval true If peer NID is on the local node.
4051 * \retval false If peer NID is not on the local node.
4053 bool LNetIsPeerLocal(lnet_nid_t nid)
4055 struct lnet_net *net;
4059 cpt = lnet_net_lock_current();
4060 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4061 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4062 if (ni->ni_nid == nid) {
4063 lnet_net_unlock(cpt);
4068 lnet_net_unlock(cpt);
4072 EXPORT_SYMBOL(LNetIsPeerLocal);
4075 * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4076 * Note that all interfaces share a same PID, as requested by LNetNIInit().
4078 * \param index Index of the interface to look up.
4079 * \param id On successful return, this location will hold the
4080 * struct lnet_process_id ID of the interface.
4082 * \retval 0 If an interface exists at \a index.
4083 * \retval -ENOENT If no interface has been found.
4086 LNetGetId(unsigned int index, struct lnet_process_id *id)
4089 struct lnet_net *net;
4093 LASSERT(the_lnet.ln_refcount > 0);
4095 cpt = lnet_net_lock_current();
4097 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4098 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4102 id->nid = ni->ni_nid;
4103 id->pid = the_lnet.ln_pid;
4109 lnet_net_unlock(cpt);
4112 EXPORT_SYMBOL(LNetGetId);
4114 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4115 struct lnet_process_id __user *ids, int n_ids)
4117 struct lnet_handle_eq eqh;
4118 struct lnet_handle_md mdh;
4119 struct lnet_event event;
4120 struct lnet_md md = { NULL };
4124 const signed long a_long_time = cfs_time_seconds(60);
4125 struct lnet_ping_buffer *pbuf;
4126 struct lnet_process_id tmpid;
4133 /* n_ids limit is arbitrary */
4134 if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4138 * if the user buffer has more space than the lnet_interfaces_max
4139 * then only fill it up to lnet_interfaces_max
4141 if (n_ids > lnet_interfaces_max)
4142 n_ids = lnet_interfaces_max;
4144 if (id.pid == LNET_PID_ANY)
4145 id.pid = LNET_PID_LUSTRE;
4147 pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4151 /* NB 2 events max (including any unlink event) */
4152 rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4154 CERROR("Can't allocate EQ: %d\n", rc);
4155 goto fail_ping_buffer_decref;
4158 /* initialize md content */
4159 md.start = &pbuf->pb_info;
4160 md.length = LNET_PING_INFO_SIZE(n_ids);
4161 md.threshold = 2; /* GET/REPLY */
4163 md.options = LNET_MD_TRUNCATE;
4167 rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4169 CERROR("Can't bind MD: %d\n", rc);
4173 rc = LNetGet(LNET_NID_ANY, mdh, id,
4174 LNET_RESERVED_PORTAL,
4175 LNET_PROTO_PING_MATCHBITS, 0, false);
4178 /* Don't CERROR; this could be deliberate! */
4179 rc2 = LNetMDUnlink(mdh);
4182 /* NB must wait for the UNLINK event below... */
4184 timeout = a_long_time;
4188 /* MUST block for unlink to complete */
4190 blocked = cfs_block_allsigs();
4192 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4195 cfs_restore_sigs(blocked);
4197 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4198 (rc2 <= 0) ? -1 : event.type,
4199 (rc2 <= 0) ? -1 : event.status,
4200 (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4202 LASSERT(rc2 != -EOVERFLOW); /* can't miss anything */
4204 if (rc2 <= 0 || event.status != 0) {
4205 /* timeout or error */
4206 if (!replied && rc == 0)
4207 rc = (rc2 < 0) ? rc2 :
4208 (rc2 == 0) ? -ETIMEDOUT :
4212 /* Ensure completion in finite time... */
4214 /* No assertion (racing with network) */
4216 timeout = a_long_time;
4217 } else if (rc2 == 0) {
4218 /* timed out waiting for unlink */
4219 CWARN("ping %s: late network completion\n",
4222 } else if (event.type == LNET_EVENT_REPLY) {
4226 } while (rc2 <= 0 || !event.unlinked);
4230 CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4237 LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4239 rc = -EPROTO; /* if I can't parse... */
4242 CERROR("%s: ping info too short %d\n",
4243 libcfs_id2str(id), nob);
4247 if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4248 lnet_swap_pinginfo(pbuf);
4249 } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4250 CERROR("%s: Unexpected magic %08x\n",
4251 libcfs_id2str(id), pbuf->pb_info.pi_magic);
4255 if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4256 CERROR("%s: ping w/o NI status: 0x%x\n",
4257 libcfs_id2str(id), pbuf->pb_info.pi_features);
4261 if (nob < LNET_PING_INFO_SIZE(0)) {
4262 CERROR("%s: Short reply %d(%d min)\n",
4264 nob, (int)LNET_PING_INFO_SIZE(0));
4268 if (pbuf->pb_info.pi_nnis < n_ids)
4269 n_ids = pbuf->pb_info.pi_nnis;
4271 if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4272 CERROR("%s: Short reply %d(%d expected)\n",
4274 nob, (int)LNET_PING_INFO_SIZE(n_ids));
4278 rc = -EFAULT; /* if I segv in copy_to_user()... */
4280 memset(&tmpid, 0, sizeof(tmpid));
4281 for (i = 0; i < n_ids; i++) {
4282 tmpid.pid = pbuf->pb_info.pi_pid;
4283 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4284 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4287 rc = pbuf->pb_info.pi_nnis;
4290 rc2 = LNetEQFree(eqh);
4292 CERROR("rc2 %d\n", rc2);
4295 fail_ping_buffer_decref:
4296 lnet_ping_buffer_decref(pbuf);
4301 lnet_discover(struct lnet_process_id id, __u32 force,
4302 struct lnet_process_id __user *ids, int n_ids)
4304 struct lnet_peer_ni *lpni;
4305 struct lnet_peer_ni *p;
4306 struct lnet_peer *lp;
4307 struct lnet_process_id *buf;
4311 int max_intf = lnet_interfaces_max;
4315 id.nid == LNET_NID_ANY)
4318 if (id.pid == LNET_PID_ANY)
4319 id.pid = LNET_PID_LUSTRE;
4322 * if the user buffer has more space than the max_intf
4323 * then only fill it up to max_intf
4325 if (n_ids > max_intf)
4328 buf_size = n_ids * sizeof(*buf);
4330 LIBCFS_ALLOC(buf, buf_size);
4334 cpt = lnet_net_lock_current();
4335 lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4342 * Clearing the NIDS_UPTODATE flag ensures the peer will
4343 * be discovered, provided discovery has not been disabled.
4345 lp = lpni->lpni_peer_net->lpn_peer;
4346 spin_lock(&lp->lp_lock);
4347 lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4348 /* If the force flag is set, force a PING and PUSH as well. */
4350 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4351 spin_unlock(&lp->lp_lock);
4352 rc = lnet_discover_peer_locked(lpni, cpt, true);
4356 /* Peer may have changed. */
4357 lp = lpni->lpni_peer_net->lpn_peer;
4358 if (lp->lp_nnis < n_ids)
4359 n_ids = lp->lp_nnis;
4363 while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4364 buf[i].pid = id.pid;
4365 buf[i].nid = p->lpni_nid;
4370 lnet_net_unlock(cpt);
4373 if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4379 lnet_peer_ni_decref_locked(lpni);
4381 lnet_net_unlock(cpt);
4383 LIBCFS_FREE(buf, buf_size);
4389 * Retrieve peer discovery status.
4391 * \retval 1 if lnet_peer_discovery_disabled is 0
4392 * \retval 0 if lnet_peer_discovery_disabled is 1
4395 LNetGetPeerDiscoveryStatus(void)
4397 return !lnet_peer_discovery_disabled;
4399 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);